🍁MMAPLE🍁 - a semi-supervised meta learning framework - Meta Model Agnostic Pseudo Label Learning

MMAPLE is a a semi-supervised meta learning framework to address these challenges by effectively exploring out-of-distribution (OOD) unlabeled data. The power of MMAPLE is demonstrated in multiple applications: predicting OOD drug-target interactions, hidden human metabolite-enzyme interactions, and understudied interspecies microbiome metabolite-human receptor interactions, where chemicals in unseen data are dramatically different from those in training data. MMAPLE achieves 11% to 242% improvement in the prediction-recall on multiple OOD benchmarks over baseline models. Using MMAPLE, we reveal novel interspecies metabolite-protein interactions that are validated by bioactivity assays and fill in missing links in microbiome-human interactions. MMAPLE is a general framework to explore previously unrecognized biological domains beyond the reach of present experimental and computational techniques.

read the paper here

This work is licensed under a Creative Commons Attribution 4.0 International License.

Architecture

Data download

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Prerequisites

To run this project, you will need to have the following installed on your local machine:

• Python 3.7 or higher
• Pytorch 1.10 or higher (with CUDA)
• Pytorch geometric
• Rdkit

Installing

1. Clone this repository onto your local machine using the following command:

   git clone https://github.com/AdorableYoyo/DESSML
   

2. Create a new Conda environment with Python 3.9 ( 3.7 or higher) by running the following command:

   conda create -n DESSML python=3.9
   

3. Activate the new Conda environment by running the following command:

   conda activate DESSML
   

4. Install the required Python packages in the Conda environment using the following command:

   bash requirments.sh
   

Stage 1 and Stage 2 model training/evaluation : Notes on Usage of Different Experiments

This README provides instructions on how to run both Stage 1 and Stage 2 experiments, incorporating semi-supervised training with unlabeled data. There are three experiments outlined below, along with the necessary file paths and instructions for cross-validation.

Experiment 1: OOD DTI

Stage 1:

To run Experiment 1 in Stage 1, follow these steps:

1. Execute the command :

   bash run_scripts/stage1_sup.sh
   

   with the following path : --train_datapath Data/DTI/chembl_train1.tsv
   --dev_datapath Data/DTI/chembl_dev1.tsv
   --test_datapath Data/DTI/DTI_test_x22102.tsv \

2. Repeat the above steps three times, each time using a different fold (1,2,3) for cross-validation.

3. After obtaining the scores for each fold, take the average as the final score.

4. This stage will output saved model in the /experiments_logs for each run.

5. For the purpose of reproducibility, you can skip stage1 since the trained models were saved in trained_model/

Stage 2:

To run Experiment 1 in Stage 2 with semi-supervised training, follow these additional steps:

1. Execute the command

   bash run_scripts/stage2_meta_student.sh
   

with the following parameters: --ul_datapath Data/DTI/DTI_target_sam_unlabeled_x53502.tsv --albert_checkpoint Data/trained_model/stage1_trained/DTI_fold_1_DISAE_model_state_dict_0.pth

Note: Data/trained_model/stage2_trained/DTI_fold_1_ours.pth is from the stage 1. It was renamed from something like : experiment_logs/exp2023-xxxxxxxx /epoch_xx/student_state_dict.pth \

1. Repeat the above steps three times, using the ALBERT checkpoint corresponding to each fold (1,2,3) for cross-validation.
2. Take the average score across the three repetitions as the final score.

Experiment 2: Hidden human MPI

Stage 1:

To run Experiment 2 Hidden human MPI in Stage 1, follow these steps:

1. Execute the command :

   bash run_scripts/stage1_sup.sh
   

   with the following path :

   --train_datapath Data/ChEMBL29/all_Chembl29.tsv
   --dev_datapath Data/HMDB/Feb_13_23_dev_test/dev_47.tsv
   --test_datapath Data/HMDB/Feb_13_23_dev_test/test_47.tsv \

2. Repeat the above steps three times, each time using a different fold (27, 37, and 47) for cross-validation.

3. After obtaining the scores for each fold, take the average as the final score.

4. This stage will output saved model in the /experiments_logs for each run.

5. For the purpose of reproducibility, you can skip stage1 since the trained models were saved in trained_model/

Stage 2:

To run Experiment 2 in Stage 2 with semi-supervised training, follow these additional steps:

1. Execute the command

   bash run_scripts/stage2_meta_student.sh
   

with the following parameters: --ul_datapath Data/new_unlabeled_data_2023/hmdb_target_sample.tsv
--albert_checkpoint Data/trained_model/stage1_trained/DISAE_tr_chembl_te_hmdb_47.pth

Note: Data/trained_model/stage1_trained/DISAE_tr_chembl_te_hmdb_47.pth is from the stage 1. It was renamed from something like : experiment_logs/exp2023-xxxxxxxx /epoch_xx/student_state_dict.pth \ \

1. Repeat the above steps three times, using the ALBERT checkpoint corresponding to each fold (27, 37, and 47) for cross-validation.
2. Take the average score across the three repetitions as the final score.

Experiment3: Zero-shot microbiome-human MPI

Stage 1:

To run Experiment 1 in Stage 1, follow these steps:

1. Execute the command :

   bash run_scripts/stage1_sup.sh
   

   with the following path :

   --train_datapath Data/Combined/activities/combined_all/train_300.tsv
   --dev_datapath Data/Combined/activities/combined_all/dev_300.tsv
   --test_datapath Data/TestingSetFromPaper/activities_nolipids.txt \

2. Repeat the above steps three times, each time using a different fold (100, 200, and 300) for cross-validation.

3. After obtaining the scores for each fold, take the average as the final score.

4. This stage will output saved model in the /experiments_logs for each run.

5. For the purpose of reproducibility, you can skip stage1 since the trained models were saved in trained_model/

Stage 2:

To run Experiment 1 in Stage 2 with semi-supervised training, follow these additional steps:

1. Execute the command

   bash run_scripts/stage2_meta_student.sh
   

with the following parameters: --ul_datapath Data/new_unlabeled_data_2023/all_gpcr_and_other_proteins_all_chem_pairs.tsv --albert_checkpoint Data/trained_model/stage1_trained/DISAE_tr_combined_te_paper_300.pth

1. Repeat the above steps three times, using the ALBERT checkpoint corresponding to each fold(100, 200, and 300) for cross-validation.
2. Take the average score across the three repetitions as the final score.

Model inference

To use the trained model to predict on desired dataset, simply run :

bash run_scripts/prediction.sh

• -state_dict_path Data/trained_model/[xxxxx desired trained model ] --datapath [desired data to be predicted]
• — get_embeddings could be added to get chemical embeddings instead of the final prediction.

Author and citation

• You Wu, Li Xie, Yang Liu, Lei Xie

  @article{wu2024semi,
  title={Semi-supervised meta-learning elucidates understudied molecular interactions},
  author={Wu, You and Xie, Li and Liu, Yang and Xie, Lei},
  journal={Communications Biology},
  volume={7},
  number={1},
  pages={1104},
  year={2024},
  publisher={Nature Publishing Group UK London}